Introduction

Ground Truth Solutions analysed survey data collected by the Zambian Red Cross in early 2023, covering questions around trust as part of the Trust Index project.

Sampling and Coverage

The sampling employed a purposive sampling approach focused on the surrounding areas of 20 pre-selected health facilities, where the Zambian Red Cross (ZRCS) has operational presence. This approach was carried out in two provinces, with two districts selected within each province. The sample was stratified based on health center.

It is important to note that this non-probability design does not allow for inferences to be made about the larger population in the selected districts, provinces, or beyond. Therefore, all the data should be regarded as indicative, and ABC strongly advises against presenting this data as representative of the entire country of Zambia, as the non-probability sampling focused on a small portion of the country.

The two provinces included in the study represent less than 25% of the overall population, and the four districts represent less than 5% of the total population of Zambia.

However, a post-stratification technique can be employed to ensure that the weighted sample corresponds to the population in the provinces according to certain parameters.

Therefore, our initial investigation focuses on a few demographic parameters of our sample and compares them to the overall population in the province.

Overall, it is not entirely clear how the health centers were selected. Additionally, sampling in the surroundings of health centers where the ZRCS operates can introduce bias to the sample.

Demographic data

For the following demographic data, we show the full data set including people who indicated to have donated to the ZRCS as well as a sub-set of the sample, that did not donate. In the survey, participants were asked whether they had ever donated to the ZRCS. It is important to note that although a significant number of respondents claimed to have made donations, the ZRCS has confirmed that they have not received any donations.

Age and gender

In terms of age and gender, the overall sample shows minor deviations from the actual population. There is a slight over-representation of females, while younger individuals are slightly under-represented.

When considering the survey respondents who did not indicate any donations to the ZRCS, the deviations from the actual population are slightly larger. However, the age distribution of the overall population is still fairly well represented in the sample.

Education

Although we have encountered challenges in accessing reliable education data, a preliminary analysis suggests that the available education data aligns with expected values for different levels of education. Specifically, we have observed that the prevalence of individuals with advanced university degrees does not appear to be disproportionately high in our dataset than in the general population.

Employment

The available data on unemployment in Zambia presents some contradictions. According to World Bank data, the labor force participation rate in 2022 was reported to be 61%. However, data from the Zambian Ministry of Labour and Social Security for 2020 indicates a lower rate of 35%.

To establish an approximation, we utilized the average employment-to-population ratio of 20.2% based on the Ministry’s data for the Eastern and Southern provinces. It is worth noting that the working age population in these two provinces is nearly identical. Additionally, it is important to consider that the working age population encompasses individuals aged 15 and above, while our sample only includes individuals aged 18 and above.

Comparing our findings to the World Bank data, it appears that the people interviewed in our sample have a significantly higher unemployment rate than the national average. In 2021, the country’s unemployment rate was reported to be 6%, and the labor force participation rate was 61% according to the World Bank’s data.

Nevertheless, it is worth noting that the Ministry of Labour and Social Security data aligns more closely with the percentage of employed individuals in our sample, which we then used in our analysis.

Distribution of mean scores for values and trust per demographic questions

By examining the average raw scores for trusting behaviors and value questions, we can generate various breakdowns based on the demographic variables provided. These breakdowns encompass both the overall sample and the subset of individuals who reported not having donated to the ZRCS.

Trust

Values

Upon analyzing the data, we have observed that there are not significant variations in the results based on age, gender, or location when examining the overall dataset (left-hand side). However, the most substantial differences arise when considering variables related to previous interactions with the ZRCS. Specifically, individuals who have volunteered, requested support, donated, or currently receive support tend to rate both trust and value questions higher compared to the rest of the sample. These differences range between 0.2 and 0.3 on the 0-3 scale, indicating an approximate 10% deviation.

These findings suggest that our overall results are influenced by the demographic distribution of our sample concerning these previously mentioned variables. As the survey participants were sampled in proximity to health centers where higher rates of volunteering, support requests, and ongoing support may be prevalent, it is reasonable to expect that, all else being equal, trust and value ratings for the overall district would be lower than in our sample.

Furthermore, when focusing solely on “non-donors,” we also observe substantial differences, with variations of up to 30% between the two provinces.

Weighting

To address the deviation of demographic parameters from the overall population, we have utilized a technique called raking. The raking process adjusts the results based on several variables to ensure that our sample reflects the distribution of these variables in the overall population. Here are the variables we considered for raking:

Additionally, we made assumptions for the upper bounds of volunteer and aid recipient percentages since no specific data was available. We estimated these rates to be 10% of the population.

Using an appropriate package in R to conduct the raking, we obtained the following results:

The line graph provides further evidence that the survey results are indeed sensitive to the inclusion of “donors” as well as the proportion of the population that volunteered or requested aid. These findings align with the previous observations from the bar chart breakdowns. As a result, we highly recommend presenting this data not as representative of Zambia as a whole but specifically as data gathered from the vicinity of ZRCS-run health centers in two provinces of Zambia. This clarification accurately reflects the limitations of the sample and avoids misrepresentation of the data as representative of the entire country.

Weighted Results

Raked data by gender, age groups and employment, considering only “non-donors” (scenario 2)

Here we are presenting the weighted data obtained through the raking process, which takes into account variables such as gender, age groups, and employment, while considering only the responses from “non-donors.” It is important to note that this data represents the vicinity of ZRCS-run health centers in two provinces of Zambia and should be considered indicative rather than representative of the entire country.

Results per beneficiaries, volunteers and others

Beneficiaries Volunteers Others
Respondents 97 63 412

Raked data by gender, age groups and employment and estimated rates of volunteers and aid requesters, considering only “non-donors” (scenario 4)

Here we are presenting the weighted data obtained through the raking process based on gender, age groups, employment, and estimated rates of volunteers and aid requesters. However, we want to emphasize that we do not recommend using these data points for presenting results. Instead, this serves as one possible way to “extrapolate” the data while ensuring an upper bound on the share of people in the weighted sample who volunteered and requested aid.

The purpose of including this information is to highlight the sensitivity of the results to the demographics of the sample in relation to these parameters. The large differences observed in this data compared to the previous dataset demonstrate how the results can vary based on the assumption that there is a larger proportion of non-volunteers and non-aid requesters who would answer in a similar manner as the non-volunteers and non-aid requesters in the vicinity of the health centers.

This data provides an insight into how the results at the district level could appear if all other factors remain constant and we assume a larger share of non-volunteers and non-aid requesters who align their responses with the non-volunteers and non-aid requesters in the proximity of the health centers.